@article{4595,
  author = {Zhijiang Lan},
  title = {Preserving Conservation Cultural Materials with AI: A Human Pose Estimation Approach},
  journal = {International Journal of Web Applications},
  year = {2025},
  volume = {17},
  number = {4},
  doi = {https://doi.org/10.6025/ijwa/2025/17/4/163-170},
  url = {https://www.dline.info/ijwa/fulltext/v17n4/ijwav17n4_3.pdf},
  abstract = {This paper explores how deep learning can be leveraged to preserve and promote traditional Chinese martial
arts, which are a vital part of China's intangible cultural heritage. Historically transmitted through master
apprentice relationships, these arts now face challenges due to societal modernization and demographic
shifts. The author proposes using deep learning models particularly convolutional neural networks (CNNs)
to capture, analyze, and reconstruct martial arts movements through human pose estimation. A novel
model, IPN (Involution Pose Estimation Net), built on Simple Baselines and Involution mechanisms, is introduced
to identify key body joints from video data with high accuracy. The study utilizes datasets like NTURGB+
D and UTD-MHAD, though it acknowledges their limitations for martial arts specific actions, highlighting
the need for a dedicated Chinese martial arts motion database. Evaluation metrics such as tMPJPE (time
aware Mean Per Joint Position Error) are adapted to assess pose accuracy over time. Experimental results
demonstrate effective movement recognition in 'Youth Serial Boxing,' with promising convergence and
accuracy. The research underscores deep learning's potential not only in digitizing martial arts but also in
enabling broader cultural preservation in the intelligent era. Future work includes expanding the model to
other martial arts styles and improving dataset specificity to enhance training and generalization.},
}